文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > 互联网 > Deep Belief Networks for Image Denoising
侵权投诉

Deep Belief Networks for Image Denoising

onthevaluesoflastlayernodes.Subsequently,ifweeliminatetheeffectsofnodespresentingthenoise,thereconstructedimagewillbenoiseless.Fromnowon,wecalledthenodespresentingnoiseandthenodespresentingimagecontent“noisenodes”and“imagenodes”,respectively.Therestofthepaperisorganizedasfollows:InSection2webrie ydescribeDeepLearning.InSection3wedescribethelearningprocess.Section4isourexperimentalresultsand nally,weconcludethepaperinSection5.

2DeepLearning

RestrictedBoltzmannMachines(RBMs)arethebuildingblocksofDBNs.Hence,inthissection rstwebrie ydescribeRBMsandthenwillexploreDBNs.

2.1RestrictedBoltzmannMachines

BoltzmannMachines(BMs)andRestrictedBoltzmannMachines(RBMs)wereintroducedin1980s.Buttheyattractedmoreattentionsince2006afterHintonetal.paper[8].HeshowedthataverypowerfulneuralnetworkcanbemadebystackingRBMs.RBMsareakindofMarkovRandomFields(MRF)whichhaveatwo-layerstructure.Onelayeriscalledvisibleandanotheriscalledhiddenlayer.Theyarerestrictedtohavenovisible-visibleor

Deep Belief Networks for Image Denoising

hidden-hiddenconnectionandcon-nectionsareinterlayer.AgraphicaldepictionofRBMisshowninFigure1.

Figure1:StructureofaRestrictedBoltzmannMachine.

Ajointcon guration,(v,h)ofvisibleandhiddenunitshasanenergygivenby[9]:

E(v,h)= aivi bjhj vihjwi,j

i∈visiblej∈hiddeni,j(1)

wherevi,hjarethebinarystatesofvisibleunitiandhiddenunitjandai,bjaretheirbiasesandwijistheweightbetweenthem.Thenetworkassignsaprobabilitytoeverypossiblepairofavisibleandahiddenvectorviathisenergyfunction[10]:

e E(v,h)

p(v,h)= E(v,h)v,he(2)

Theprobabilitythatthenetworkassignstoavisiblevector,v,isgivenbysummingoverallpossiblehiddenvectors: E(v,h)ep(v)=h

E(v,h)(3)ev,h

logp(v))=<vihj>data <vihj>model ωi,j(4)

wheretheanglebracketsareusedtodenoteexpectationsunderthedistributionspeciesbythesub-scriptthatfollows.Thisleadstoaverysimplelearningruleforperformingstochasticsteepestascentinthelogprobabilityofthetrainingdata:

ωij= (<vihj>data <vihj>model)

(5)

上一页第2页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email